test: avoid of too long running sql-tap tests

The test sql-tap/gh-3083-ephemeral-unref-tuples.test.lua took
too long runtime and failed to finish in timeout limit of 2
minutes when it runs near the end of the running queue:

No output during 120 seconds. Will abort after 120 seconds without output. List of workers not reporting the status:
- 012_sql-tap [sql-tap/gh-3083-ephemeral-unref-tuples.test.lua, vinyl] at var/012_sql-tap/gh-3083-ephemeral-unref-tuples.result:0
Test hung! Result content mismatch:
[File does not exists: sql-tap/gh-3083-ephemeral-unref-tuples.result]
[Main process] No output from workers. It seems that we hang. Send SIGKILL to workers; exiting...

Due to suggestion from Konstantin Osipov at gh-3845:
https://github.com/tarantool/tarantool/issues/3845#issue-385735959

The following tests:
  gh-3083-ephemeral-unref-tuples.test.lua
  gh-3332-tuple-format-leak.test.lua
where set to run only on memtx configuration.

Also the test:
  gh-3083-ephemeral-unref-tuples.test.lua
was removed from the test-run 'fragile' option flaky list.

Closed #4128

test/sql-tap/engine.cfg

@@ -14,6 +14,12 @@
     "sort.test.lua": {
         "memtx": {"engine": "memtx"}
     },
+    "gh-3083-ephemeral-unref-tuples.test.lua": {
+        "memtx": {"engine": "memtx"}
+    },
+    "gh-3332-tuple-format-leak.test.lua": {
+        "memtx": {"engine": "memtx"}
+    },
     "gh-4077-iproto-execute-no-bind.test.lua": {},
     "*": {
         "memtx": {"engine": "memtx"},

test/sql-tap/suite.ini

@@ -29,6 +29,5 @@ long_run = gh-3332-tuple-format-leak.test.lua, gh-3083-ephemeral-unref-tuples.te
 config = engine.cfg
 show_reproduce_content = False
 pretest_clean = True
-fragile = gh-3083-ephemeral-unref-tuples.test.lua ; gh-4128
-          gh-4077-iproto-execute-no-bind.test.lua ; gh-4459
+fragile = gh-4077-iproto-execute-no-bind.test.lua ; gh-4459
           selectG.test.lua                        ; gh-4458